Nonstationary vector AR modeling of wireless channels

We propose a novel long-term parametric channel model for wireless channels that do not satisfy the usual assumption of wide-sense stationary uncorrelated scattering (WSSUS). Our non-WSSUS channel model is based on an innovations system representation that models the channel taps jointly as nonstationary vector autoregressive (AR) process, leading to a highly parsimonious description of the channel statistics. We also present a Yule-Walker type approach for estimating the parameters of this vector time-frequency AR model. Numerical results involving a simulated channel and real-world measurements illustrate the usefulness of our approach.